Financial services are failing to successfully implement artificial intelligence, European fintech executives claim, even as evidence mounts that the hyped technology will increase productivity and reduce costs.
Fear of job losses, regulatory concerns and institutional inertia are some of the factors preventing bankers from fully embracing the systems underlying products like ChatGPT.
“The big banks will certainly not adopt this [the technology] as fast as any fintech,” said Tom Blomfield, co-founder of Monzo and group partner at Silicon Valley start-up incubator Y Combinator. Generative AI, however, will make banks “more efficient and able to offer the same products at a lower cost.”
A Capgemini survey shows that only 6 percent of retail banks are willing to implement AI in their businesses at scale. However, McKinsey estimates that this could add up to $340 billion in value to the global banking sector annually, equivalent to around 4.7 percent of the sector’s total revenues.
Many say the technology, with its ability to answer questions and analyze vast amounts of text and numerical data in seconds, has the power to drive down costs across the industry. Still, there are concerns that the disruption will lead to job losses.
“People don’t understand that it’s there as a productivity tool,” says Nasir Zubairi, CEO of fintech accelerator Luxembourg House of Financial Technology. “They still genuinely believe it’s going to take their jobs away.”
He added: “Traditional banks are fundamentally analogue by design, and converting from analogue to digital has always been difficult.”
Speaking at the Financial Times’ TNW tech conference this month, Zubairi cited anti-money laundering controls as an example. Institutions typically hire employees to comb through spreadsheets looking for unusual activity.
He said that when he showed one institution how this could be improved with a bespoke AI model, which he estimated could immediately save up to €450,000 a year in salary, it was rejected.
“People don’t like to fire people,” he added. “They want to protect the function of their work and if they have to fire people within their team who perform these tasks, they are also potentially threatened because management or their power is also eroded in some way.”
Central banks have recently been urged to “up their game” with AI, according to the Bank for International Settlements. The bank said the technology could deliver productivity gains but also pose risks, such as providing false information and being vulnerable to hacking.
A common problem with large language models, the technology behind most generative AI products, is their tendency to “hallucinate,” to present inaccuracies as fact. They are also known to generate information based on the data they are trained on, leading to concerns about sensitive or secure information.
“There is not necessarily a rejection of [AI]“But there is hesitation,” said Wincie Wong, head of digital at NatWest, who called for the technology’s risks, ethics and vulnerabilities to be assessed before it is implemented. “Ultimately, we are one of the big banks and many customers keep their data and finances safe with us. We need to respect that.”
Customer service is one of the areas most disrupted by AI tools, which can converse and respond to queries in a human-like manner. Digital banks have been using machine learning to sort online queries for more than a decade, often directing customers to a live customer service representative.
However, bots using LLM can understand a wider range of questions, regardless of how they are phrased. In addition, they can execute decisions such as ordering a bank card, eliminating the need for human intervention.
“I really think it will eliminate the vast majority of customer service jobs” in “the next 12 months to the next five years,” Monzo’s Blomfield said.
Many banks and fintechs, including Klarna and NatWest, are already using AI chatbots for customer service. NatWest’s Wong said they had made huge strides with generative AI in their service AI Cora, which received more than 11 million chats in the year, more than half of which required no human intervention. In 2017, the service was receiving 1,000 chats per month and needed intervention.
Swedish fintech Klarna said its AI assistant could do the work of 700 customer service agents and resolve queries in under two minutes, up from 11 minutes previously. As a result, the company expects to save $40 million in customer service costs this year.
However, Wong said training the models to be nuanced was crucial to its success, understanding that a change of address could have emotional undertones, like a family death, for example.
“Understanding the psychology behind it was really important and if you don’t get it right you can, to put it bluntly, piss off a lot of customers,” she added.
Banks also had to tread carefully as they rolled out the emerging technology while adhering to the industry’s strict compliance rules and navigating an unfamiliar regulatory environment.
In a landmark ruling from 2022, a Dutch court ruled in favor of neobank Bunq after it sued the Dutch Central Bank for banning the use of AI in money laundering checks.
Regulators last month lifted restrictions on German fintech N26 after it beefed up its controls. For years, the bank had capped new customer sign-ups because of poor anti-money laundering controls and faced millions of euros in fines for persistently late filing of suspicious activity reports.
Carina Kozole, Chief Risk Officer at N26, said the company has worked closely with regulators to develop an AI model to assess whether a new customer is a criminal, which has reduced the number of cases on the platform by 90 percent.
“If we don’t embrace AI in the industry, we won’t be here in a few years,” she added. “We need to show the benefits and how we can become compliant when we use AI.”